Neural Architecture Search (NAS) has attracted increasing attention in recent years because of its capability to design neural networks automatically. Among them, differential NAS approaches such as DARTS, have gained popularity for search efficiency. However, they still suffer from three main issues, that are, the weak stability due to the performance collapse, the poor generalization ability of the searched architectures, and the inferior robustness to different kinds of proxies (i.e., computationally reduced search configurations). To solve the search stability and searched architecture's generalization problems, a simple-but-effective regularization method, termed as Beta-Decay, is proposed to regularize the DARTS-based NAS searching process (referred as $\beta$β-DARTS). Specifically, Beta-Decay regularization can impose constraints to keep the value and variance of activated architecture parameters from being too large, thereby ensuring fair competition among architecture parameters and making the supernet less sensitive to the impact of input on the operation set. In-depth theoretical analyses on how it works and why it works are provided, and comprehensive experiments on a variety of search spaces and datasets validate that Beta-Decay regularization can help to stabilize the searching process and make the searched network more transferable across different datasets. To address the proxy robustness problem, we first benchmark differentiable NAS methods under a wide range of proxy data, proxy channels, proxy layers, and proxy epochs, since the robustness of NAS under different kinds of proxies has not been explored before. We then conclude some interesting findings and find that $\beta$β-DARTS always achieves the best result among all compared NAS methods under almost all proxy settings. We further introduce the novel flooding regularization to the weight optimization of $\beta$β-DARTS (termed as Bi-level regularization), and experimentally and theoretically verify its effectiveness for improving the proxy robustness of differentiable NAS.

Dynamic surface reconstruction requires not only photorealistic rendering but also temporally stable geometry. In this paper, we present DynaSurfGS, a dynamic surface reconstruction framework built on planar Gaussian splatting with a new temporal path consistency constraint. The key idea is to treat temporal coherence as a path-invariant property of deformation. For a given canonical primitive, its future state should be consistent, regardless of whether it is obtained by direct prediction or temporal propagation through an intermediate state. This trajectory-level formulation regularizes deformation in motion space rather than only supervising per-frame geometry, thereby suppressing temporal instability at its source. Built upon planar dynamic Gaussians and complemented by local geometric supervision, our framework produces both high-quality rendering and geometrically stable dynamic surfaces. Experiments demonstrate that DynaSurfGS achieves strong rendering performance while producing more coherent dynamic surfaces over time.

3D Gaussian Splatting (3DGS) has shown promising performance in novel view synthesis. Previous methods adapt it to obtaining surfaces of either individual 3D objects or within limited scenes. In this paper, we make the first attempt to tackle the challenging task of large-scale scene surface reconstruction. This task is particularly difficult due to the high GPU memory consumption, different levels of details for geometric representation, and noticeable inconsistencies in appearance. To this end, we propose GigaGS, the first work for high-quality surface reconstruction for large-scale scenes using 3DGS. GigaGS first applies a partitioning strategy based on the mutual visibility of spatial regions, which effectively grouping cameras for parallel processing. To enhance the quality of the surface, we also propose novel multi-view photometric and geometric consistency constraints based on Level-of-Detail representation. In doing so, our method can reconstruct detailed surface structures. Comprehensive experiments are conducted on various datasets. The consistent improvement demonstrates the superiority of GigaGS.

We propose GoalFlow, an end-to-end autonomous driving method for generating high-quality multimodal trajectories. In autonomous driving scenarios, there is rarely a single suitable trajectory. Recent methods have increasingly focused on modeling multimodal trajectory distributions. However, they suffer from trajectory selection complexity and reduced trajectory quality due to high trajectory divergence and inconsistencies between guidance and scene information. To address these issues, we introduce GoalFlow, a novel method that effectively constrains the generative process to produce high-quality, multimodal trajectories. To resolve the trajectory divergence problem inherent in diffusion-based methods, GoalFlow constrains the generated trajectories by introducing a goal point. GoalFlow establishes a novel scoring mechanism that selects the most appropriate goal point from the candidate points based on scene information. Furthermore, GoalFlow employs an efficient generative method, Flow Matching, to generate multimodal trajectories, and incorporates a refined scoring mechanism to select the optimal trajectory from the candidates. Our experimental results, validated on the Navsim[7], demonstrate that GoalFlow achieves state-of-the-art performance, delivering robust multimodal trajectories for autonomous driving. GoalFlow achieved PDMS of 90.3, significantly surpassing other methods. Compared with other diffusion-policy-based methods, our approach requires only a single denoising step to obtain excellent performance. The code is available at https://github.com/YvanYin/GoalFlow.

Learning from multiple domains is a primary factor that influences the generalization of a single unified robot system. In this paper, we aim to learn the trajectory prediction model by using broad out-of-domain data to improve its performance and generalization ability. Trajectory model is designed to predict any-point trajectories in the current frame given an instruction and can provide detailed control guidance for robotic policy learning. To handle the diverse out-of-domain data distribution, we propose a sparsely-gated MoE (Top-1 gating strategy) architecture for trajectory model, coined as Tra-MoE. The sparse activation design enables good balance between parameter cooperation and specialization, effectively benefiting from large-scale out-of-domain data while maintaining constant FLOPs per token. In addition, we further introduce an adaptive policy conditioning technique by learning 2D mask representations for predicted trajectories, which is explicitly aligned with image observations to guide action prediction more flexibly. We perform extensive experiments on both simulation and real-world scenarios to verify the effectiveness of Tra-MoE and adaptive policy conditioning technique. We also conduct a comprehensive empirical study to train Tra-MoE, demonstrating that our Tra-MoE consistently exhibits superior performance compared to the dense baseline model, even when the latter is scaled to match Tra-MoE’s parameter count.

In this paper, we introduce an innovative vector quantization based action tokenizer built upon the largest-scale action trajectory dataset to date, leveraging over 100 times more data than previous approaches. This extensive dataset enables our tokenizer to capture rich spatiotemporal dynamics, resulting in a model that not only accelerates inference but also generates smoother and more coherent action outputs. Once trained, the tokenizer can be seamlessly adapted to a wide range of downstream tasks in a zero-shot manner, from short-horizon reactive behaviors to long-horizon planning. A key finding of our work is that the domain gap between synthetic and real action trajectories is marginal, allowing us to effectively utilize a vast amount of synthetic data during training without compromising real-world performance. To validate our approach, we conducted extensive experiments in both simulated environments and on real robotic platforms. The results demonstrate that as the volume of synthetic trajectory data increases, the performance of our tokenizer on downstream tasks improves significantly-most notably, achieving up to a 30% higher success rate on two real-world tasks in long-horizon scenarios. These findings highlight the potential of our action tokenizer as a robust and scalable solution for real-time embodied intelligence systems, paving the way for more efficient and reliable robotic control in diverse application domains.Project website: https://xiaoxiao0406.github.io/vqvla.github.io

Recently, 3D assets created via reconstruction and generation have matched the quality of manually crafted assets, highlighting their potential for replacement. However, this potential is largely unrealized because these assets always need to be converted to meshes for 3D industry applications, and the meshes produced by current mesh extraction methods are significantly inferior to Artist-Created Meshes (AMs), i.e., meshes created by human artists. Specifically, current mesh extraction methods rely on dense faces and ignore geometric features, leading to inefficiencies, complicated post-processing, and lower representation quality. To address these issues, we introduce MeshAnything, a model that treats mesh extraction as a generation problem, producing AMs aligned with specified shapes. By converting 3D assets in any 3D representation into AMs, MeshAnything can be integrated with various 3D asset production methods, thereby enhancing their application across the 3D industry. The architecture of MeshAnything comprises a VQ-VAE and a shape-conditioned decoder-only transformer. We first learn a mesh vocabulary using the VQ-VAE, then train the shape-conditioned decoder-only transformer on this vocabulary for shape-conditioned autoregressive mesh generation. Our extensive experiments show that our method generates AMs with hundreds of times fewer faces, significantly improving storage, rendering, and simulation efficiencies, while achieving precision comparable to previous methods.

Spatial intelligence is the ability of a machine to perceive, reason, and act in three dimensions within space and time. Recent advancements in large-scale auto-regressive models have demonstrated remarkable capabilities across various reasoning tasks. However, these models often struggle with fundamental aspects of spatial reasoning, particularly in answering questions like "Where am I?" and "What will I see?". While some attempts have been done, existing approaches typically treat them as separate tasks, failing to capture their interconnected nature. In this paper, we present **G**enerative **S**patial **T**ransformer (GST), a novel auto-regressive framework that jointly addresses spatial localization and view prediction. Our model simultaneously estimates the camera pose from a single image and predicts the view from a new camera pose, effectively bridging the gap between spatial awareness and visual prediction. The proposed innovative camera tokenization method enables the model to learn the joint distribution of 2D projections and their corresponding spatial perspectives in an auto-regressive manner. This unified training paradigm demonstrates that joint optimization of pose estimation and novel view synthesis leads to improved performance in both tasks, for the first time, highlighting the inherent relationship between spatial awareness and visual prediction.

Sora unveils the potential of scaling Diffusion Transformer (DiT) for generating photorealistic images and videos at arbitrary resolutions, aspect ratios, and durations, yet it still lacks sufficient implementation details. In this paper, we introduce the Lumina-T2X family -- a series of Flow-based Large Diffusion Transformers (Flag-DiT) equipped with zero-initialized attention, as a simple and scalable generative framework that can be adapted to various modalities, e.g., transforming noise into images, videos, multi-view 3D objects, or audio clips conditioned on text instructions. By tokenizing the latent spatial-temporal space and incorporating learnable placeholders such as |[nextline]| and |[nextframe]| tokens, Lumina-T2X seamlessly unifies the representations of different modalities across various spatial-temporal resolutions. Advanced techniques like RoPE, KQ-Norm, and flow matching enhance the stability, flexibility, and scalability of Flag-DiT, enabling models of Lumina-T2X to scale up to 7 billion parameters and extend the context window to 128K tokens. This is particularly beneficial for creating ultra-high-definition images with our Lumina-T2I model and long 720p videos with our Lumina-T2V model. Remarkably, Lumina-T2I, powered by a 5-billion-parameter Flag-DiT, requires only 35% of the training computational costs of a 600-million-parameter naive DiT (PixArt-alpha), indicating that increasing the number of parameters significantly accelerates convergence of generative models without compromising visual quality. Our further comprehensive analysis underscores Lumina-T2X's preliminary capability in resolution extrapolation, high-resolution editing, generating consistent 3D views, and synthesizing videos with seamless transitions. All code and checkpoints of Lumina-T2X are released at https://github.com/Alpha-VLLM/Lumina-T2X to further foster creativity, transparency, and diversity in the generative AI community.

Neural implicit reconstruction via volume rendering has demonstrated its effectiveness in recovering dense 3D surfaces. However, it is non-trivial to simultaneously recover meticulous geometry and preserve smoothness across regions with differing characteristics. To address this issue, previous methods typically employ geometric priors, which are often constrained by the performance of the prior models. In this paper, we propose ND-SDF, which learns a Normal Deflection field to represent the angular deviation between the scene normal and the prior normal. Unlike previous methods that uniformly apply geometric priors on all samples, introducing significant bias in accuracy, our proposed normal deflection field dynamically learns and adapts the utilization of samples based on their specific characteristics, thereby improving both the accuracy and effectiveness of the model. Our method not only obtains smooth weakly textured regions such as walls and floors but also preserves the geometric details of complex structures. In addition, we introduce a novel ray sampling strategy based on the deflection angle to facilitate the unbiased rendering process, which significantly improves the quality and accuracy of intricate surfaces, especially on thin structures. Consistent improvements on various challenging datasets demonstrate the superiority of our method.

Video depth estimation has long been hindered by the scarcity of consistent and scalable ground truth data, leading to inconsistent and unreliable results. In this paper, we introduce Depth Any Video, a model that tackles the challenge through two key innovations. First, we develop a scalable synthetic data pipeline, capturing real-time video depth data from diverse virtual environments, yielding 40,000 video clips of 5-second duration, each with precise depth annotations. Second, we leverage the powerful priors of generative video diffusion models to handle real-world videos effectively, integrating advanced techniques such as rotary position encoding and flow matching to further enhance flexibility and efficiency. Unlike previous models, which are limited to fixed-length video sequences, our approach introduces a novel mixed-duration training strategy that handles videos of varying lengths and performs robustly across different frame rates—even on single frames. At inference, we propose a depth interpolation method that enables our model to infer high-resolution video depth across sequences of up to 150 frames. Our model outperforms all previous generative depth models in terms of spatial accuracy and temporal consistency. The code and model weights are open-sourced.

In this paper, we introduce SPA, a novel representation learning framework that emphasizes the importance of 3D spatial awareness in embodied AI. Our approach leverages differentiable neural rendering on multi-view images to endow a vanilla Vision Transformer (ViT) with intrinsic spatial understanding. We present the most comprehensive evaluation of embodied representation learning to date, covering 268 tasks across 8 simulators with diverse policies in both single-task and language-conditioned multi-task scenarios. The results are compelling: SPA consistently outperforms more than 10 state-of-the-art representation methods, including those specifically designed for embodied AI, vision-centric tasks, and multi-modal applications, while using less training data. Furthermore, we conduct a series of real-world experiments to confirm its effectiveness in practical scenarios. These results highlight the critical role of 3D spatial awareness for embodied representation learning. Our strongest model takes more than 6000 GPU hours to train and we are committed to open-sourcing all code and model weights to foster future research in embodied representation learning.

Video generation techniques have made remarkable progress, promising to be the foundation of interactive world exploration.However, existing video generation datasets are not well-suited for world exploration training as they suffer from some limitations: limited locations, short duration, static scenes, and a lack of annotations about exploration and the world.In this paper, we introduce Sekai (meaning "world" in Japanese), a high-quality first-person view worldwide video dataset with rich annotations for world exploration. It consists of over 5,000 hours of walking or drone view (FPV and UVA) videos from over 100 countries and regions across 750 cities. We develop an efficient and effective toolbox to collect, pre-process and annotate videos with location, scene, weather, crowd density, captions, and camera trajectories.Comprehensive analyses and experiments demonstrate the dataset’s scale, diversity, annotation quality, and effectiveness for training video generation models.We believe Sekai will benefit the area of video generation and world exploration, and motivate valuable applications.

This paper's primary objective is to develop a robust generalist perception model capable of addressing multiple tasks under constraints of computational resources and limited training data. We leverage text-to-image diffusion models pre-trained on billions of images and successfully introduce our DICEPTION, a visual generalist model. Exhaustive evaluations demonstrate that DICEPTION effectively tackles diverse perception tasks, even achieving performance comparable to SOTA single-task specialist models. Specifically, we achieve results on par with SAM-vit-h using only 0.06% of their data (e.g., 600K vs.\ 1B pixel-level annotated images). We designed comprehensive experiments on architectures and input paradigms, demonstrating that the key to successfully re-purposing a single diffusion model for multiple perception tasks lies in maximizing the preservation of the pre-trained model's prior knowledge. Consequently, DICEPTION can be trained with substantially lower computational costs than conventional models requiring training from scratch. Furthermore, adapting DICEPTION to novel tasks is highly efficient, necessitating fine-tuning on as few as 50 images and approximately 1% of its parameters. Finally, we demonstrate that a subtle application of classifier-free guidance can improve the model's performance on depth and normal estimation. We also show that pixel-aligned training, as is characteristic of perception tasks, significantly enhances the model's ability to preserve fine details. DICEPTION offers valuable insights and presents a promising direction for the development of advanced diffusion-based visual generalist models.

Geometry plays a significant role in monocular 3D object detection. It can be used to estimate object depth by using the perspective projection between object's physical size and 2D projection in the image plane, which can introduce mathematical priors into deep models. However, this projection process also introduces error amplification, where the error of the estimated height is amplified and reflected into the projected depth. It leads to unreliable depth inferences and also impairs training stability. To tackle this problem, we propose a novel Geometry Uncertainty Propagation Network (GUPNet++) by modeling geometry projection in a probabilistic manner. This ensures depth predictions are well-bounded and associated with a reasonable uncertainty. The significance of introducing such geometric uncertainty is two-fold: (1). It models the uncertainty propagation relationship of the geometry projection during training, improving the stability and efficiency of the end-to-end model learning. (2). It can be derived to a highly reliable confidence to indicate the quality of the 3D detection result, enabling more reliable detection inference. Experiments show that the proposed approach not only obtains (state-of-the-art) SOTA performance in image-based monocular 3D detection but also demonstrates superiority in efficacy with a simplified framework. The code and model will be released at https://github.com/SuperMHP/GUPNet_Plus.

Human-centric perception tasks, e.g., pedestrian detection, skeleton-based action recognition, and pose estimation, have wide industrial applications, such as metaverse and sports analysis. There is a recent surge to develop human-centric foundation models that can benefit a broad range of human-centric perception tasks. While many human-centric foundation models have achieved success, they did not explore 3D and vision-language tasks for human-centric and required task-specific finetuning. These limitations restrict their application to more downstream tasks and situations. To tackle these problems, we present Hulk, the first multimodal human-centric generalist model, capable of addressing 2D vision, 3D vision, skeleton-based, and vision-language tasks without task-specific finetuning. The key to achieving this is condensing various task-specific heads into two general heads, one for discrete representations, e.g., languages, and the other for continuous representations, e.g., location coordinates. The outputs of two heads can be further stacked into four distinct input and output modalities. This uniform representation enables Hulk to treat diverse human-centric tasks as modality translation, integrating knowledge across a wide range of tasks. Comprehensive evaluations of Hulk on 12 benchmarks covering 8 human-centric tasks demonstrate the superiority of our proposed method, achieving state-of-the-art performance in 11 benchmarks.

Residual networks have shown great success and become indispensable in recent deep neural network models. In this work, we aim to re-investigate the training process of residual networks from a novel perspective of loafing, and further propose a new training scheme as well as three improved strategies for boosting residual networks beyond their performance limits. Previous research has suggested that residual networks can be considered as ensembles of shallow networks, which implies that the final performance of a residual network is influenced by a group of subnetworks. Furthermore, we identify a previously overlooked problem, where subnetworks within a residual network are prone to exert less effort when working as part of a group compared to working alone. We define this problem as network loafing. Since network loafing may inevitably cause the sub-par performance of the residual network, we propose a novel training scheme called stimulative training, which randomly samples a residual subnetwork and calculates the KL divergence loss between the sampled subnetwork and the given residual network for extra supervision. In order to unleash the potential of stimulative training, we further propose three simple-yet-effective strategies, including a novel KL- loss that only aligns the network logits direction, random smaller inputs for subnetworks, and inter-stage sampling rules. Comprehensive experiments and analysis verify the effectiveness of stimulative training as well as its three improved strategies. For example, the proposed method can boost the performance of ResNet50 on ImageNet to 80.5% Top1 accuracy without using any extra data, model, trick, or changing the structure. With only uniform augment, the performance can be further improved to 81.0% Top1 accuracy, better than the best training recipes provided by Timm library and PyTorch official version. We also verify its superiority on various typical models, datasets, and tasks and give some theoretical analysis. As such, we advocate utilizing the proposed method as a general and next-generation technology to train residual networks.

In contrast to numerous NLP and 2D vision foundational models, training a 3D foundational model poses considerably greater challenges. This is primarily due to the inherent data variability and diversity of downstream tasks. In this paper, we introduce a novel universal 3D pre-training framework designed to facilitate the acquisition of efficient 3D representations. Considering that informative 3D features should encode rich geometry and appearance cues that can be utilized to render realistic images, we propose to learn 3D representations by differentiable neural rendering. We train a 3D backbone with a volumetric neural renderer by comparing the rendered with the real images. Notably, our pre-trained encoder can be seamlessly applied to various downstream tasks. These tasks include semantic challenges like 3D detection and segmentation, which involve scene understanding, and non-semantic tasks like 3D reconstruction and image synthesis, which focus on geometry and visuals. They span both indoor and outdoor scenarios. We also illustrate the capability of pre-training a 2D backbone using the proposed methodology, surpassing conventional pre-training methods by a large margin. For the first time, PonderV2 achieves state-of-the-art performance on 11 indoor and outdoor benchmarks, implying its effectiveness.

Reconstructing urban street scenes is crucial due to its vital role in applications such as autonomous driving and urban planning. These scenes are characterized by long, narrow camera trajectories, occlusion, complex object relationships, and sparse data across multiple scales. Despite recent advancements, existing surface reconstruction methods, which are primarily designed for object-centric scenarios, struggle to adapt effectively to the unique characteristics of street scenes. To address this challenge, we introduce StreetSurfGS, the first method to employ Gaussian Splatting specifically tailored for scalable urban street scene surface reconstruction. StreetSurfGS utilizes a planar-based octree representation and segmented training to reduce memory costs, accommodate unique camera characteristics, and improve scalability. Additionally, to mitigate depth inaccuracies caused by object overlap, we propose a guided smoothing strategy within regularization to eliminate inaccurate boundary points and outliers. Furthermore, to address sparse views and multi-scale challenges, we use a dual-step matching strategy that leverages adjacent and long-term information. Extensive experiments validate the efficacy of StreetSurfGS in both novel view synthesis and surface reconstruction.

As Vision Transformers (ViTs) become increasingly popular in various vision tasks, one may question:if a new training scheme for ViTs exists that can improve performance without increasing training and inference computation cost?In this paper, we affirmatively answer this question with a novel Sparse-to-Dense (S2D) training scheme. Specifically, we decouple the training and inference phases of ViTs. During training, we replace some Feed-Forward Network (FFN) layers of ViTs with computationally efficient RUP-Mixture-of-FFN (RUP-MoF) layers, each comprising multiple FFN experts and allocating tokens to experts via Random Uniform Partition (RUP). Furthermore, an additional Experts Weights Averaging (EWA) update is performed specifically on these RUP-MoF layers after each gradient update. After training, we convert each RUP-MoF layer back to a single FFN layer by averaging the experts, transforming the training-time sparse model back to the original dense ViT model for inference. We further provide theoretical analysis to illustrate why and how it works. Comprehensive experiments across various 2D and 3D vision tasks, ViT architectures and datasets validate the effectiveness and generalization ability of the proposed S2D training scheme. Besides, we show that, S2D training scheme can also be applied to improve the performance of Transformer-based language models, and EWA update technique can also significantly improve the effectiveness of classic Mixture-of-Experts on various 2D vision small-scale datasets and 3D vision tasks.

The pretrain-finetune paradigm has achieved great success in NLP and 2D image fields because of the high-quality representation ability and transferability of their pretrained models. However, pretraining such a strong model is difficult in the 3D point cloud field due to the limited amount of point cloud sequences. This paper introduces Efficient Point Cloud Learning (EPCL), an effective and efficient point cloud learner for directly training high-quality point cloud models with a frozen CLIP transformer. Our EPCL connects the 2D and 3D modalities by semantically aligning the image features and point cloud features without paired 2D-3D data. Specifically, the input point cloud is divided into a series of local patches, which are converted to token embeddings by the designed point cloud tokenizer. These token embeddings are concatenated with a task token and fed into the frozen CLIP transformer to learn point cloud representation. The intuition is that the proposed point cloud tokenizer projects the input point cloud into a unified token space that is similar to the 2D images. Comprehensive experiments on 3D detection, semantic segmentation, classification and few-shot learning demonstrate that the CLIP transformer can serve as an efficient point cloud encoder and our method achieves promising performance on both indoor and outdoor benchmarks. In particular, performance gains brought by our EPCL are 19.7 AP50 on ScanNet V2 detection, 4.4 mIoU on S3DIS segmentation and 1.2 mIoU on SemanticKITTI segmentation compared to contemporary pretrained models. Code is available at \url{https://github.com/XiaoshuiHuang/EPCL}.

Recent research understands the residual networks from a new perspective of the implicit ensemble model. From this view, previous methods such as stochastic depth and stimulative training have further improved the performance of the residual network by sampling and training of its subnets. However, they both use the same supervision for all subnets of different capacities and neglect the valuable knowledge generated by subnets during training. In this manuscript, we mitigate the significant knowledge distillation gap caused by using the same kind of supervision and advocate leveraging the subnets to provide diverse knowledge. Based on this motivation, we propose a group knowledge based training framework for boosting the performance of residual networks. Specifically, we implicitly divide all subnets into hierarchical groups by subnet-in-subnet sampling, aggregate the knowledge of different subnets in each group during training, and exploit upper-level group knowledge to supervise lower-level subnet group. Meanwhile, we also develop a subnet sampling strategy that naturally samples larger subnets, which are found to be more helpful than smaller subnets in boosting performance for hierarchical groups. Compared with typical subnet training and other methods, our method achieves the best efficiency and performance trade-offs on multiple datasets and network structures. The code is at https://github.com/tsj-001/AAAI24-GKT.

In this work, we introduce Context-Aware MultiModal Learner (CaMML), for tuning large multimodal models (LMMs).CaMML, a lightweight module, is crafted to seamlessly integrate multimodal contextual samples into large models, thereby empowering the model to derive knowledge from analogous, domain-specific, up-to-date information and make grounded inferences.Importantly, CaMML is highly scalable and can efficiently handle lengthy multimodal context examples owing to its hierarchical design.Based on CaMML, we have developed two multimodal models, CaMML-7B and CaMML-13B, that have shown exceptional performance across an array of benchmark datasets for multimodal tasks.Remarkably, CaMML-13B achieves the state-of-the-art performance on over ten widely recognized multimodal benchmark datasets, surpassing LLaVA-1.5 (13B) with a noticeable margin, without integration of any external resources.Moreover, we have conducted extensive ablative studies to inspect the inner workings of CaMML and performed qualitative analyses to showcase its effectiveness in handling real-world challenging cases.Code and models are available at: https://github. com/amazon-science/camml.

This paper is not motivated to seek innovation within the attention mechanism. Instead, it focuses on overcoming the existing trade-offs between accuracy and efficiency within the context of point cloud processing, leveraging the power of scale. Drawing inspiration from recent advances in 3D large-scale representation learning, we recognize that model performance is more influenced by scale than by intricate design. Therefore, we present Point Transformer V3 (PTv3), which prioritizes simplicity and efficiency over the accuracy of certain mechanisms that are minor to the over-all performance after scaling, such as replacing the precise neighbor search by KNN with an efficient serialized neighbor mapping of point clouds organized with specific patterns. This principle enables significant scaling, expanding the receptive field from 16 to 1024 points while remaining efficient (a 3 x increase in processing speed and a 10 x improvement in memory efficiency compared with its pre-decessor, PTv2). PTv3 attains state-of-the-art results on over 20 downstream tasks that span both indoor and out-door scenarios. Further enhanced with multi-dataset joint training, PTv3 pushes these results to a higher level.

Training deep models for LiDAR semantic segmentation is challenging due to the inherent sparsity of point clouds. Utilizing temporal data is a natural remedy against the spar-sity problem as it makes the input signal denser. However, previous multi-frame fusion algorithms fall short in utilizing sufficient temporal information due to the memory constraint, and they also ignore the informative temporal images. To fully exploit rich information hidden in long-term temporal point clouds and images, we present the Temporal Aggre-gation Network, termed TASeg. Specifically, we propose a Temporal LiDAR Aggregation and Distillation (TLAD) algorithm, which leverages historical priors to assign dif-ferent aggregation steps for different classes. It can largely reduce memory and time overhead while achieving higher accuracy. Besides, TLAD trains a teacher injected with gt priors to distill the model, further boosting the performance. To make full use of temporal images, we design a Temporal Image Aggregation and Fusion (TIAF) module, which can greatly expand the camera FOVand enhance the present features. Temporal LiDAR points in the camera FOV are used as mediums to transform temporal image features to the present coordinate for temporal multi-modal fusion. Moreover, we develop a Static-Moving Switch Augmentation (SMSA) algorithm, which utilizes sufficient temporal information to enable objects to switch their motion states freely, thus greatly increasing static and moving training samples. Our TASeg ranks 1st††the date of CVPR deadline, i.e., 2023-11-18 07:59 AM UTC. on three challenging tracks, i.e., SemanticKITTI single-scan track, multi-scan track and nuScenes LiDAR segmentation track, strongly demonstrating the superiority of our method. Codes are available at https://github.com/LittlePey/TASeg.

Utilizing pretrained 2D large-scale generative models, recent works are capable of generating high-quality novel views from a single in-the-wild image. However, due to the lack of information from multiple views, these works encounter difficulties in generating controllable novel views. In this paper, we present DreamComposer, a flexible and scalable framework that can enhance existing view-aware diffusion models by injecting multi-view conditions. Specifically, DreamComposer first uses a view-aware 3D lifting module to obtain 3D representations of an object from multiple views. Then, it renders the latent features of the target view from 3D representations with the multi-view feature fusion module. Finally the target view features extracted from multi-view inputs are injected into a pretrained diffusion model. Experiments show that DreamComposer is compatible with state-of-the-art diffusion models for zero-shot novel view synthesis, further enhancing them to generate high-fidelity novel view images with multi-view conditions, ready for controllable 3D object reconstruction and various other applications.

In the context of autonomous driving, the significance of effective feature learning is widely acknowledged. While conventional 3D self-supervised pretraining methods have shown widespread success, most methods follow the ideas originally designed for 2D images. In this paper, we present UniPAD, a novel self-supervised learning paradigm applying 3D volumetric differentiable rendering. UniPAD implicitly encodes 3D space, facilitating the reconstruction of continuous 3D shape structures and the intricate appear-ance characteristics of their 2D projections. The flexibil-ity of our method enables seamless integration into both 2D and 3D frameworks, enabling a more holistic compre-hension of the scenes. We manifest the feasibility and effectiveness of UniPAD by conducting extensive experiments on various 3D perception tasks. Our method significantly improves lidar-, camera-, and lidar-camera-based baseline by 9.1, 7.7, and 6.9 NDS, respectively. Notably, our pretraining pipeline achieves 73.2 NDS for 3D object detection and 79.4 mIoU for 3D semantic segmentation on the nuScenes validation set, achieving state-of-the-art results in comparison with previous methods.

The Segment-Anything Model (SAM) stands as a foundational framework for image segmentation. While it exhibits remarkable zero-shot generalization in typical scenarios, its advantage diminishes when applied to specialized domains like medical imagery and remote sensing. To address this limitation, this paper introduces Conv-LoRA, a simple yet effective parameter-efficient fine-tuning approach. By integrating ultra-lightweight convolutional parameters into Low-Rank Adaptation (LoRA), Conv-LoRA can inject image-related inductive biases into the plain ViT encoder, further reinforcing SAM’s local prior assumption. Notably, Conv-LoRA not only preserves SAM’s extensive segmentation knowledge but also revives its capacity of learning high-level image semantics, which is constrained by SAM’s foreground-background segmentation pretraining. Comprehensive experimentation across diverse benchmarks spanning multiple domains underscores Conv-LoRA’s superiority in adapting SAM to real-world semantic segmentation tasks.

The success of pretrain-finetune paradigm brings about the release of numerous model weights. In this case, merging models finetuned on different tasks to enable a single model with multi-task capabilities is gaining increasing attention for its practicability. Existing model merging methods usually suffer from (1) significant performance degradation or (2) requiring tuning by additional data or training. In this paper, we rethink and analyze the existing model merging paradigm. We discover that using a single model's weights can hardly simulate all the models' performance. To tackle this issue, we propose Elect, Mask & Rescale-Merging (EMR-Merging). We first (a) elect a unified model from all the model weights and then (b) generate extremely lightweight task-specific modulators, including masks and rescalers, to align the direction and magnitude between the unified model and each specific model, respectively. EMR-Merging is tuning-free, thus requiring no data availability or any additional training while showing impressive performance. We find that EMR-Merging shows outstanding performance compared to existing merging methods under different classical and newly-established settings, including merging different numbers of vision models (up to 30), NLP models, PEFT models, and multi-modal models.

Signed Distance Function (SDF)-based volume rendering has demonstrated significant capabilities in surface reconstruction. Although promising, SDF-based methods often fail to capture detailed geometric structures, resulting in visible defects. By comparing SDF-based volume rendering to density-based volume rendering, we identify two main factors within the SDF-based approach that degrade surface quality: SDF-to-density representation and geometric regularization. These factors introduce challenges that hinder the optimization of the SDF field. To address these issues, we introduce NeuRodin, a novel two-stage neural surface reconstruction framework that not only achieves high-fidelity surface reconstruction but also retains the flexible optimization characteristics of density-based methods. NeuRodin incorporates innovative strategies that facilitate transformation of arbitrary topologies and reduce artifacts associated with density bias. Extensive evaluations on the Tanks and Temples and ScanNet++ datasets demonstrate the superiority of NeuRodin, showing strong reconstruction capabilities for both indoor and outdoor environments using solely posed RGB captures. Project website: https://open3dvlab.github.io/NeuRodin/

Diffusion-based methods have achieved remarkable achievements in 2D image or 3D object generation, however, the generation of 3D scenes and even $360^{\circ}$ images remains constrained, due to the limited number of scene datasets, the complexity of 3D scenes themselves, and the difficulty of generating consistent multi-view images. To address these issues, we first establish a large-scale panoramic video-text dataset containing millions of consecutive panoramic keyframes with corresponding panoramic depths, camera poses, and text descriptions. Then, we propose a novel text-driven panoramic generation framework, termed DiffPano, to achieve scalable, consistent, and diverse panoramic scene generation. Specifically, benefiting from the powerful generative capabilities of stable diffusion, we fine-tune a single-view text-to-panorama diffusion model with LoRA on the established panoramic video-text dataset. We further design a spherical epipolar-aware multi-view diffusion model to ensure the multi-view consistency of the generated panoramic images. Extensive experiments demonstrate that DiffPano can generate scalable, consistent, and diverse panoramic images with given unseen text descriptions and camera poses.

In robot learning, the observation space is crucial due to the distinct characteristics of different modalities, which can potentially become a bottleneck alongside policy design. In this study, we explore the influence of various observation spaces on robot learning, focusing on three predominant modalities: RGB, RGB-D, and point cloud. We introduce OBSBench, a benchmark comprising two simulators and 125 tasks, along with standardized pipelines for various encoders and policy baselines. Extensive experiments on diverse contact-rich manipulation tasks reveal a notable trend: point cloud-based methods, even those with the simplest designs, frequently outperform their RGB and RGB-D counterparts. This trend persists in both scenarios: training from scratch and utilizing pre-training. Furthermore, our findings demonstrate that point cloud observations often yield better policy performance and significantly stronger generalization capabilities across various geometric and visual conditions. These outcomes suggest that the 3D point cloud is a valuable observation modality for intricate robotic tasks. We also suggest that incorporating both appearance and coordinate information can enhance the performance of point cloud methods. We hope our work provides valuable insights and guidance for designing more generalizable and robust robotic models.

The effectiveness of spectral-spatial feature learning is crucial for the hyperspectral image (HSI) classification task. Diffusion models, as a new class of groundbreaking generative models, have the ability to learn both contextual semantics and textual details from the distinct timestep dimension, enabling the modeling of complex spectral-spatial relations in HSIs. However, existing diffusion-based HSI classification methods only utilize manually selected single-timestep single-stage features, limiting the full exploration and exploitation of rich contextual semantics and textual information hidden in the diffusion model. To address this issue, we propose a novel diffusion-based feature learning framework that explores Multi-Timestep Multi-Stage Diffusion features for HSI classification for the first time, called MTMSD. Specifically, the diffusion model is first pretrained with unlabeled HSI patches to mine the connotation of unlabeled data, and then is used to extract the multi-timestep multi-stage diffusion features. To effectively and efficiently leverage multi-timestep multi-stage features, two strategies are further developed. One strategy is class & timestep-oriented multi-stage feature purification module with the inter-class and inter-timestep prior for reducing the redundancy of multi-stage features and alleviating memory constraints. The other one is selective timestep feature fusion module with the guidance of global features to adaptively select different timestep features for integrating texture and semantics. Both strategies facilitate the generality and adaptability of the MTMSD framework for diverse patterns of different HSI data. Extensive experiments are conducted on four public HSI datasets, and the results demonstrate that our method outperforms state-of-the-art methods for HSI classification, especially on the challenging Houston 2018 dataset. The codes are available at https://github.com/zjyaccount/MTMSD.

Existing image captioning methods are under the assumption that the training and testing data are from the same domain or that the data from the target domain (i.e., the domain that testing data lie in) are accessible. However, this assumption is invalid in real-world applications where the data from the target domain is inaccessible. In this paper, we introduce a new setting called Domain Generalization for Image Captioning (DGIC), where the data from the target domain is unseen in the learning process. We first construct a benchmark dataset for DGIC, which helps us to investigate models' domain generalization (DG) ability on unseen domains. With the support of the new benchmark, we further propose a new framework called language-guided semantic metric learning (LSML) for the DGIC setting. Experiments on multiple datasets demonstrate the challenge of the task and the effectiveness of our newly proposed benchmark and LSML framework.

Masked Modeling (MM) has demonstrated widespread success in various vision challenges, by reconstructing masked visual patches. Yet, applying MM for large-scale 3D scenes remains an open problem due to the data sparsity and scene complexity. The conventional random masking paradigm used in 2D images often causes a high risk of ambiguity when recovering the masked region of 3D scenes. To this end, we propose a novel informative-preserved reconstruction, which explores local statistics to discover and preserve the representative structured points, effectively enhancing the pretext masking task for 3D scene understanding. Integrated with a progressive reconstruction manner, our method can concentrate on modeling regional geometry and enjoy less ambiguity for masked reconstruction. Besides, such scenes with progressive masking ratios can also serve to self-distill their intrinsic spatial consistency, requiring to learn the consistent representations from unmasked areas. By elegantly combining informative-preserved reconstruction on masked areas and consistency self-distillation from unmasked areas, a unified framework called MM-3DScene is yielded. We conduct comprehensive experiments on a host of downstream tasks. The consistent improvement (e.g., +6.1% [email protected] on object detection and +2.2% mIoU on semantic segmentation) demonstrates the superiority of our approach.

Recent Transformer-based 3D object detectors learn point cloud features either from point- or voxel-based representations. However, the former requires time-consuming sampling while the latter introduces quantization errors. In this paper, we present a novel Point-Voxel Transformer for single-stage 3D detection (PVT-SSD) that takes advantage of these two representations. Specifically, we first use voxel-based sparse convolutions for efficient feature encoding. Then, we propose a Point-Voxel Transformer (PVT) module that obtains long-range contexts in a cheap manner from voxels while attaining accurate positions from points. The key to associating the two different representations is our introduced input-dependent Query Initialization module, which could efficiently generate reference points and content queries. Then, PVT adaptively fuses long-range contextual and local geometric information around reference points into content queries. Further, to quickly find the neighboring points of reference points, we design the Virtual Range Image module, which generalizes the native range image to multi-sensor and multi-frame. The experiments on several autonomous driving benchmarks verify the effectiveness and efficiency of the proposed method. Code will be available.

Despite the tremendous progress of Masked Autoencoders (MAE) in developing vision tasks such as image and video, exploring MAE in large-scale 3D point clouds remains challenging due to the inherent irregularity. In contrast to previous 3D MAE frameworks, which either design a complex decoder to infer masked information from maintained regions or adopt sophisticated masking strategies, we instead propose a much simpler paradigm. The core idea is to apply a Generative Decoder for MAE (GD-MAE) to automatically merges the surrounding context to restore the masked geometric knowledge in a hierarchical fusion manner. In doing so, our approach is free from introducing the heuristic design of decoders and enjoys the flexibility of exploring various masking strategies. The corresponding part costs less than 12% latency compared with conventional methods, while achieving better performance. We demonstrate the efficacy of the proposed method on several large-scale benchmarks: Waymo, KITTI, and ONCE. Consistent improvement on downstream detection tasks illustrates strong robustness and generalization capability. Not only our method reveals state-of-the-art results, but remarkably, we achieve comparable accuracy even with 20% of the labeled data on the Waymo dataset. Code will be released.

We propose a novel approach to self-supervised learning of point cloud representations by differentiable neural rendering. Motivated by the fact that informative point cloud features should be able to encode rich geometry and appearance cues and render realistic images, we train a point-cloud encoder within a devised point-based neural renderer by comparing the rendered images with real images on massive RGB-D data. The learned point-cloud encoder can be easily integrated into various downstream tasks, including not only high-level tasks like 3D detection and segmentation but also low-level tasks like 3D reconstruction and image synthesis. Extensive experiments on various tasks demonstrate the superiority of our approach compared to existing pre-training methods.

In this paper, we come up with a simple yet effective approach for instance segmentation on 3D point cloud with strong robustness. Previous top-performing methods for this task adopt a bottom-up strategy, which often involves various inefficient operations or complex pipelines, such as grouping over-segmented components, introducing heuristic post-processing steps, and designing complex loss functions. As a result, the inevitable variations of the instances sizes make it vulnerable and sensitive to the values of pre-defined hyper-parameters. To this end, we instead propose a novel pipeline that applies dynamic convolution to generate instance-aware parameters in response to the characteristics of the instances. The representation capability of the parameters is greatly improved by gathering homogeneous points that have identical semantic categories and close votes for the geometric centroids. Instances are then decoded via several simple convolution layers, where the parameters are generated depending on the input. In addition, to introduce a large context and maintain limited computational overheads, a light-weight transformer is built upon the bottleneck layer to capture the long-range dependencies. With the only post-processing step, non-maximum suppression (NMS), we demonstrate a simpler and more robust approach that achieves promising performance on various datasets: ScanNetV2, S3DIS, and PartNet. The consistent improvements on both voxel- and point-based architectures imply the effectiveness of the proposed method. Code is available at: https://git.io/DyCo3D.

Long-tail distribution is widely spread in real-world applications. Due to the extremely small ratio of instances, tail categories often show inferior accuracy. In this paper, we find such performance bottleneck is mainly caused by the imbalanced gradients, which can be categorized into two parts: (1) positive part, deriving from the samples of the same category, and (2) negative part, contributed by other categories. Based on comprehensive experiments, it is also observed that the gradient ratio of accumulated positives to negatives is a good indicator to measure how balanced a category is trained. Inspired by this, we come up with a gradient-driven training mechanism to tackle the long-tail problem: re-balancing the positive/negative gradients dynamically according to current accumulative gradients, with a unified goal of achieving balance gradient ratios. Taking advantage of the simple and flexible gradient mechanism, we introduce a new family of gradient-driven loss functions, namely equalization losses. We conduct extensive experiments on a wide spectrum of visual tasks, including two-stage/single-stage long-tailed object detection (LVIS), long-tailed image classification (ImageNet-LT, Places-LT, iNaturalist), and long-tailed semantic segmentation (ADE20 K). Our method consistently outperforms the baseline models, demonstrating the effectiveness and generalization ability of the proposed equalization losses.

Point cloud completion aims to predict complete shape from its partial observation. Current approaches mainly consist of generation and refinement stages in a coarse-to-fine style. However, the generation stage often lacks robustness to tackle different incomplete variations, while the refinement stage blindly recovers point clouds without the semantic awareness. To tackle these challenges, we unify point cloud Completion by a generic Pretrain-Prompt-Predict paradigm, namely CP3. Inspired by prompting approaches from NLP, we creatively reinterpret point cloud generation and refinement as the prompting and predicting stages, respectively. Then, we introduce a concise self-supervised pretraining stage before prompting. It can effectively increase robustness of point cloud generation, by an Incompletion-Of-Incompletion (IOI) pretext task. Moreover, we develop a novel Semantic Conditional Refinement (SCR) network at the predicting stage. It can discriminatively modulate multi-scale refinement with the guidance of semantics. Finally, extensive experiments demonstrate that our CP3 outperforms the state-of-the-art methods with a large margin. code will be available at https://github.com/MingyeXu/cp3.

Compared to typical multi-sensor systems, monocular 3D object detection has attracted much attention due to its simple configuration. However, there is still a significant gap between LiDAR-based and monocular-based methods. In this paper, we find that the ill-posed nature of monocular imagery can lead to depth ambiguity. Specifically, objects with different depths can appear with the same bounding boxes and similar visual features in the 2D image. Unfortunately, the network cannot accurately distinguish different depths from such non-discriminative visual features, resulting in unstable depth training. To facilitate depth learning, we propose a simple yet effective plug-and-play module, One Bounding Box Multiple Objects (OBMO). Concretely, we add a set of suitable pseudo labels by shifting the 3D bounding box along the viewing frustum. To constrain the pseudo-3D labels to be reasonable, we carefully design two label scoring strategies to represent their quality. In contrast to the original hard depth labels, such soft pseudo labels with quality scores allow the network to learn a reasonable depth range, boosting training stability and thus improving final performance. Extensive experiments on KITTI and Waymo benchmarks show that our method significantly improves state-of-the-art monocular 3D detectors by a significant margin (The improvements under the moderate setting on KITTI validation set are 1.82 ~ 10.91% mAP in BEV and 1.18 ~ 9.36% mAP in 3D). Codes have been released at https://github.com/mrsempress/OBMO.

This paper presents an approach that reconstructs a hand-held object from a monocular video. In contrast to many recent methods that directly predict object geometry by a trained network, the proposed approach does not require any learned prior about the object and is able to recover more accurate and detailed object geometry. The key idea is that the hand motion naturally provides multiple views of the object and the motion can be reliably estimated by a hand pose tracker. Then, the object geometry can be recovered by solving a multi-view reconstruction problem. We devise an implicit neural representation-based method to solve the reconstruction problem and address the issues of imprecise hand pose estimation, relative hand-object motion, and insufficient geometry optimization for small objects. We also provide a newly collected dataset with 3D ground truth to validate the proposed approach. The dataset and code will be released at https://dihuangdh.github.io/hhor.

End-to-end text-spotting, which aims to integrate detection and recognition in a unified framework, has attracted increasing attention due to its simplicity of the two complimentary tasks. It remains an open problem especially when processing arbitrarily-shaped text instances. Previous methods can be roughly categorized into two groups: character-based and segmentation-based, which often require character-level annotations and/or complex post-processing due to the unstructured output. Here, we tackle end-to-end text spotting by presenting Adaptive Bezier Curve Network v2 (ABCNet v2). Our main contributions are four-fold: 1) For the first time, we adaptively fit arbitrarily-shaped text by a parameterized Bezier curve, which, compared with segmentation-based methods, can not only provide structured output but also controllable representation. 2) We design a novel BezierAlign layer for extracting accurate convolution features of a text instance of arbitrary shapes, significantly improving the precision of recognition over previous methods. 3) Different from previous methods, which often suffer from complex post-processing and sensitive hyper-parameters, our ABCNet v2 maintains a simple pipeline with the only post-processing non-maximum suppression (NMS). 4) As the performance of text recognition closely depends on feature alignment, ABCNet v2 further adopts a simple yet effective coordinate convolution to encode the position of the convolutional filters, which leads to a considerable improvement with negligible computation overhead. Comprehensive experiments conducted on various bilingual (English and Chinese) benchmark datasets demonstrate that ABCNet v2 can achieve state-of-the-art performance while maintaining very high efficiency. More importantly, as there is little work on quantization of text spotting models, we quantize our models to improve the inference time of the proposed ABCNet v2. This can be valuable for real-time applications. Code and model are available at: https://git.io/AdelaiDet.

In computer vision, object detection is one of most important tasks, which underpins a few instance-level recognition tasks and many downstream applications. Recently one-stage methods have gained much attention over two-stage approaches due to their simpler design and competitive performance. Here we propose a fully convolutional one-stage object detector (FCOS) to solve object detection in a per-pixel prediction fashion, analogue to other dense prediction problems such as semantic segmentation. Almost all state-of-the-art object detectors such as RetinaNet, SSD, YOLOv3, and Faster R-CNN rely on pre-defined anchor boxes. In contrast, our proposed detector FCOS is anchor box free, as well as proposal free. By eliminating the pre-defined set of anchor boxes, FCOS completely avoids the complicated computation related to anchor boxes such as calculating the intersection over union (IoU) scores during training. More importantly, we also avoid all hyper-parameters related to anchor boxes, which are often sensitive to the final detection performance. With the only post-processing non-maximum suppression (NMS), we demonstrate a much simpler and flexible detection framework achieving improved detection accuracy. We hope that the proposed FCOS framework can serve as a simple and strong alternative for many other instance-level tasks. Code is available at: git.io/AdelaiDet.

Previous top-performing approaches for point cloud in-stance segmentation involve a bottom-up strategy, which often includes inefficient operations or complex pipelines, such as grouping over-segmented components, introducing additional steps for refining, or designing complicated loss functions. The inevitable variation in the instance scales can lead bottom-up methods to become particularly sensitive to hyper-parameter values. To this end, we propose in-stead a dynamic, proposal-free, data-driven approach that generates the appropriate convolution kernels to apply in response to the nature of the instances. To make the kernels discriminative, we explore a large context by gathering homogeneous points that share identical semantic categories and have close votes for the geometric centroids. Instances are then decoded by several simple convolutional layers. Due to the limited receptive field introduced by the sparse convolution, a small light-weight transformer is also devised to capture the long-range dependencies and high-level interactions among point samples. The proposed method achieves promising results on both ScanetNetV2 and S3DIS, and this performance is robust to the particular hyper-parameter values chosen. It also improves inference speed by more than 25% over the current state-of-the-art. Code is available at: https://git.io/DyCo3D

Scene flow in 3D point clouds plays an important role in understanding dynamic environments. Although significant advances have been made by deep neural networks, the performance is far from satisfactory as only per-point translational motion is considered, neglecting the constraints of the rigid motion in local regions. To address the issue, we propose to introduce the motion consistency to force the smoothness among neighboring points. In addition, constraints on the rigidity of the local transformation are also added by sharing unique rigid motion parameters for all points within each local region. To this end, a high-order CRFs based relation module (Con-HCRFs) is deployed to explore both point-wise smoothness and region-wise rigidity. To empower the CRFs to have a discriminative unary term, we also introduce a position-aware flow estimation module to be incorporated into the Con-HCRFs. Comprehensive experiments on FlyingThings3D and KITTI show that our proposed framework (HCRF-Flow) achieves state-of-the-art performance and significantly outperforms previous approaches substantially.

Scene text detection and recognition has received increasing research attention. Existing methods can be roughly categorized into two groups: character-based and segmentation-based. These methods either are costly for character annotation or need to maintain a complex pipeline, which is often not suitable for real-time applications. Here we address the problem by proposing the Adaptive Bezier-Curve Network (\BeCan). Our contributions are three-fold: 1) For the first time, we adaptively fit oriented or curved text by a parameterized Bezier curve. 2) We design a novel BezierAlign layer for extracting accurate convolution features of a text instance with arbitrary shapes, significantly improving the precision compared with previous methods. 3) Compared with standard bounding box detection, our Bezier curve detection introduces negligible computation overhead, resulting in superiority of our method in both efficiency and accuracy. Experiments on oriented or curved benchmark datasets, namely Total-Text and CTW1500, demonstrate that \BeCan achieves state-of-the-art accuracy, meanwhile significantly improving the speed. In particular, on Total-Text, our real-time version is over 10 times faster than recent state-of-the-art methods with a competitive recognition accuracy. Code is available at \url{https://git.io/AdelaiDet}.

Both accuracy and efficiency are of significant importance to the task of semantic segmentation. Existing deep FCNs suffer from heavy computations due to a series of high-resolution feature maps for preserving the detailed knowledge in dense estimation. Although reducing the feature map resolution (i.e., applying a large overall stride) via subsampling operations (e.g., polling and convolution striding) can instantly increase the efficiency, it dramatically decreases the estimation accuracy. To tackle this dilemma, we propose a knowledge distillation method tailored for semantic segmentation to improve the performance of the compact FCNs with large overall stride. To handle the inconsistency between the features of the student and teacher network, we optimize the feature similarity in a transferred latent domain formulated by utilizing a pre-trained autoencoder. Moreover, an affinity distillation module is proposed to capture the long-range dependency by calculating the non local interactions across the whole image. To validate the effectiveness of our proposed method, extensive experiments have been conducted on three popular benchmarks: Pascal VOC, Cityscapes and Pascal Context. Built upon a highly competitive baseline, our proposed method can improve the performance of a student network by 2.5% (mIOU boosts from 70.2 to 72.7 on the cityscapes test set) and can train a better compact model with only 8% float operations (FLOPS) of a model that achieves comparable performances.

Recent semantic segmentation methods exploit encoder-decoder architectures to produce the desired pixel-wise segmentation prediction. The last layer of the decoders is typically a bilinear upsampling procedure to recover the final pixel-wise prediction. We empirically show that this oversimple and data-independent bilinear upsampling may lead to sub-optimal results. In this work, we propose a data-dependent upsampling (DUpsampling) to replace bilinear, which takes advantages of the redundancy in the label space of semantic segmentation and is able to recover the pixel-wise prediction from low-resolution outputs of CNNs. The main advantage of the new upsampling layer lies in that with a relatively lower-resolution feature map such as 1/16 or 1/32 of the input size, we can achieve even better segmentation accuracy, significantly reducing computation complexity. This is made possible by 1) the new upsampling layer's much improved reconstruction capability; and more importantly 2) the DUpsampling based decoder's flexibility in leveraging almost arbitrary combinations of the CNN encoders' features. Experiments on PASCAL VOC demonstrate that with much less computation complexity, our decoder outperforms the state-of-the-art decoder. Finally, without any post-processing, the framework equipped with our proposed decoder achieves new state-of-the-art performance on two datasets: 88.1% mIOU on PASCAL VOC with 30% computation of the previously best model; and 52.5% mIOU on PASCAL Context.

We propose a fully convolutional one-stage object detector (FCOS) to solve object detection in a per-pixel prediction fashion, analogue to semantic segmentation. Almost all state-of-the-art object detectors such as RetinaNet, SSD, YOLOv3, and Faster R-CNN rely on pre-defined anchor boxes. In contrast, our proposed detector FCOS is anchor box free, as well as proposal free. By eliminating the pre-defined set of anchor boxes, FCOS completely avoids the complicated computation related to anchor boxes such as calculating overlapping during training. More importantly, we also avoid all hyper-parameters related to anchor boxes, which are often very sensitive to the final detection performance. With the only post-processing non-maximum suppression (NMS), FCOS with ResNeXt-64x4d-101 achieves 44.7% in AP with single-model and single-scale testing, surpassing previous one-stage detectors with the advantage of being much simpler. For the first time, we demonstrate a much simpler and flexible detection framework achieving improved detection accuracy. We hope that the proposed FCOS framework can serve as a simple and strong alternative for many other instance-level tasks. Code is available at: https://tinyurl.com/FCOSv1.

Text detection and recognition in natural images have long been considered as two separate tasks that are processed sequentially. Jointly training two tasks is non-trivial due to significant differences in learning difficulties and convergence rates. In this work, we present a conceptually simple yet efficient framework that simultaneously processes the two tasks in a united framework. Our main contributions are three-fold: (1) we propose a novel text-alignment layer that allows it to precisely compute convolutional features of a text instance in arbitrary orientation, which is the key to boost the performance; (2) a character attention mechanism is introduced by using character spatial information as explicit supervision, leading to large improvements in recognition; (3) two technologies, together with a new RNN branch for word recognition, are integrated seamlessly into a single model which is end-to-end trainable. This allows the two tasks to work collaboratively by sharing convolutional features, which is critical to identify challenging text instances. Our model obtains impressive results in end-to-end recognition on the ICDAR 2015 [19], significantly advancing the most recent results [2], with improvements of F-measure from (0.54, 0.51, 0.47) to (0.82, 0.77, 0.63), by using a strong, weak and generic lexicon respectively. Thanks to joint training, our method can also serve as a good detector by achieving a new state-of-the-art detection performance on related benchmarks. Code is available at https://github.com/tonghe90/textspotter.

We present a novel single-shot text detector that directly outputs word-level bounding boxes in a natural image. We propose an attention mechanism which roughly identifies text regions via an automatically learned attentional map. This substantially suppresses background interference in the convolutional features, which is the key to producing accurate inference of words, particularly at extremely small sizes. This results in a single model that essentially works in a coarse-to-fine manner. It departs from recent FCN-based text detectors which cascade multiple FCN models to achieve an accurate prediction. Furthermore, we develop a hierarchical inception module which efficiently aggregates multi-scale inception features. This enhances local details, and also encodes strong context information, allowing the detector to work reliably on multi-scale and multi-orientation text with single-scale images. Our text detector achieves an F-measure of 77% on the ICDAR 2015 benchmark, advancing the state-of-the-art results in [18, 28]. Demo is available at: http://sstd.whuang.org/.

Recent deep learning models have demonstrated strong capabilities for classifying text and non-text components in natural images. They extract a high-level feature globally computed from a whole image component (patch), where the cluttered background information may dominate true text features in the deep representation. This leads to less discriminative power and poorer robustness. In this paper, we present a new system for scene text detection by proposing a novel text-attentional convolutional neural network (Text-CNN) that particularly focuses on extracting text-related regions and features from the image components. We develop a new learning mechanism to train the Text-CNN with multi-level and rich supervised information, including text region mask, character label, and binary text/non-text information. The rich supervision information enables the Text-CNN with a strong capability for discriminating ambiguous texts, and also increases its robustness against complicated background components. The training process is formulated as a multi-task learning problem, where low-level supervised information greatly facilitates the main task of text/non-text classification. In addition, a powerful low-level detector called contrast-enhancement maximally stable extremal regions (MSERs) is developed, which extends the widely used MSERs by enhancing intensity contrast between text patterns and background. This allows it to detect highly challenging text patterns, resulting in a higher recall. Our approach achieved promising results on the ICDAR 2013 data set, with an F-measure of 0.82, substantially improving the state-of-the-art results.